On Tue, Jan 09, 2018 at 08:29:44AM +0000, Stuart Henderson wrote: > On 2018/01/09 16:03, Jonathan Gray wrote: > > On Mon, Jan 08, 2018 at 11:42:37PM -0500, Brian Callahan wrote: > > > Attached is the heavy hammer disabling all of SSE. This leaves in MMX and > > > 3DNow! flags, as I do think it's reasonable to say that CPU that pre-date > > > those extensions are probably too slow to run anything SDL2 at a usable > > > speed. > > > > SSE/SSE2 at least is unconditionally available for amd64. There may also be > > cpuid based runtime checks for the instructions so this diff seems wrong. > > There are runtime cpuid checks. I don't think we should do anything > with this now but obviously if people run into SIGILL that could > be investigated. >
Based on this discussion, I'm gonna leave SDL2 as it is for now. I'm happy to revisit this if issues are reported that might relate to this. I also think that performance on systems with SSE/SSE2 would suffer if we did this.
